A changing pattern of childhood BMI growth during the 20th century: 70 y of data from the Fels Longitudinal Study

Abstract

Background: The BMI distribution shifted upward in the United States between the 1960s and the 1990s, but little is known about secular trends in the pattern of BMI growth, particularly earlier in the century and early in childhood.

Objective: The objective was to examine differences in BMI growth in children born in 1929–1999.

Design: BMI curves from ages 2 to 18 y were produced for 855 European-American children in the Fels Longitudinal Study born in 1929–1953, 1954–1972, and 1973–1999. Age (A_min) and BMI (BMI_min) at adiposity rebound and age (AV_max), BMI (BMIV_max), and velocity (V_max) at maximum velocity were derived; multivariable regression was used to examine whether maternal BMI, infant weight gain, and other covariates mediated the cohort effects on these traits.

Results: BMI curves showed that children born in 1973–1999 had the lowest BMI values until age 5 y but had the largest values from age 8 y onward. In adjusted models, boys and girls born in 1973–1999 had a 0.15-kg/m² per year faster V_max and a 1-kg/m² higher BMIV_max than did children of the same sex born in 1929–1953, and girls had a 0.8-y earlier A_min (P < 0.01). Maternal BMI and infant weight gain were associated with an obesity-prone pattern of BMI growth but did not account for the observed trends.

Conclusions: Shifts in the BMI growth rate around the time of pubertal initiation were apparent starting after 1973. The BMI growth curve did not increase monotonically over time; rather, children born during the obesity epidemic were characterized by lower BMI values before the adiposity rebound and by rapid subsequent BMI gain.

See corresponding editorial on page 999.

INTRODUCTION

In the United States, the BMI distribution shifted upward between the 1960s and the 1990s (1–4), such that the mean BMI between 5 and 24 y of age increased by ∼1 to 2 units [ie, weight (kg)/height (m)²] (1), and significant but smaller changes occurred earlier in childhood (2). The largest secular changes in BMI have been observed at the upper end of the distribution (5, 6), and the prevalence of children aged 6–17 y with a BMI above the 85th centile cutoff point from the National Health Examination Survey increased by 7% between the National Health Examination Survey in 1963–1970 and NHANES III in 1988–1991 (4). Although these nationally representative surveys provide invaluable information as we track the prevalence of obesity at the population level, they do not address precisely how the childhood BMI growth curve and landmarks in this curve (eg, the nadir in the BMI growth curve at ∼5 y of age marking the onset of adiposity rebound) have changed over time.

Research studies of specific populations have aimed to quantify secular trends in childhood BMI, although most have assessed trends over short periods of time in the second half of the 20th century and/or at cross-sectional ages (7–11), because data collected throughout childhood and also over a substantial period of time are rare. These publications (7–9, 11) have typically found that childhood BMI increased over time during some period after the 1970s and thus add to the findings of national surveys. They have not, however, investigated trends in the shape of the BMI growth curve throughout childhood and therefore provide little evidence of how trends in BMI develop over age and over time. Such information is useful not only to identify appropriate ages and patterns of BMI growth for targeted intervention but also to provide insight into the etiologic factors responsible (12, 13). In addition, none of the aforementioned publications have investigated the factors that may have changed over time to explain a secular trend in increasing childhood BMI.

The aims of the current study were 1) to investigate secular trends over 70 y in the 20th century in the pattern of BMI growth from 2 to 18 y of age in both boys and girls, and 2) to examine secular trends in traits that characterize individual BMI growth curves (eg, age at adiposity rebound and BMI at the adiposity rebound) and test whether these potential trends have occurred independently of infant weight gain, maternal BMI, and other potential mediating variables.

SUBJECTS AND METHODS

Sample

The sample consisted of 855 European-American participants (437 boys; 418 girls) in the Fels Longitudinal Study born between 1929 and 1999. The Fels Longitudinal Study was described in detail elsewhere (14). In brief, this study began in 1929 as a study of normative child growth and development and continues today as a study of the early-life antecedents of chronic diseases of aging. Mother-infant dyads living in Yellow Springs and other towns in southwestern Ohio have been enrolled from 1929 onward. All protocols and informed consent documents were approved by the Wright State University Institutional Review Board. Parents of minors provided written consent for their children, and minor children aged 7–17 y provided their additional assent.

Measurements

Serial weight and height measurements were collected at 6 monthly intervals from 2 to 18 y of age by using standard anthropometric methods (15). BMI was calculated when weight and height were available at the same study visit. In total, there were 15,679 BMI measurements, with an average of 18.3 per participant (range: 3–30) over an average of 13.0 y (range 3.8–16.1 y). Nonpregnant maternal height and BMI were assessed at the study visit closest to the birth date of the child, and these variables included both prepregnancy and postpartum data. Maternal age at the child's birth, gestational age at birth, birth order, and birth weight of the child were collected from birth records and maternal self-report at the child's first visit to the study center. The analysis sample of 855 children was selected on the basis of being born between 1928 and 1999 and having good serial BMI data for growth modeling (ie, minimum of 3 serial measurements). Scatter plots of BMI against age for this sample and for all Fels Longitudinal Study participants born in 1928–1999 with BMI data between 2 and 18 y of age (n = 1253) showed that the BMI distribution by age was similar for the analysis sample of 855 children compared with all those Fels Longitudinal Study children with data. In addition, there was no significant difference in the first BMI measurement (ie, closest to age 2 y), after being converted to z scores against the WHO 2006 Child Growth Standard (16), between the 855 children included in the current analysis and the 398 children not included in the analysis by using an independent-samples t test (P > 0.6).

Statistical analysis

Participants were assigned to birth year tertiles to create 3 empirical birth cohorts (cohort 1: 1928–1953, n = 287; cohort 2: 1954–1972, n = 297; cohort 3: 1973–1999, n = 271). Using xtmixed in Stata, mixed-effects regression analysis was used to model the serial BMI data for boys and girls separately. A cubic function of age (age, age², and age³) with an unstructured covariance structure fit the BMI data best for each sex. Random effects were specified for the intercept and the age and age² terms. Birth cohort was included as a main effect and also as an interaction with the age terms of the cubic function. With use of the models, the mean constant BMI curves from 2 to 18 y of age were plotted by sex and birth cohort. Estimated BMI values were calculated for each child at 2 yearly intervals, and separate ANOVA models (including a Bonferroni correction for multiple comparisons) were applied to the estimated BMI data for each sex and at each age to test for differences between cohorts.

To better describe the cohort effect on characteristics of the individual BMI growth curves, the fixed and random-effect estimates for the constant, and also the appropriate main cohort effect, were summed for each individual to calculate individual constant values (ie, β₀). The same was done for the first 2 age terms to calculate individual curve parameters (ie, β₁ and β₂). Within each cohort, the third curve parameter (ie, β₃) was the same for all individuals because a random effect could not be fitted for the age³ term. These estimates of β₀, β₁, β₂, and β₃ were then used to calculate BMI growth curve traits for each child, which are known to be associated with adulthood adiposity (17): age at adiposity rebound (A_min)⁴, BMI at adiposity rebound (BMI_min), age at maximum BMI velocity (AV_max), BMI at maximum BMI velocity (BMIV_max), and BMI velocity at maximum BMI velocity (V_max). The equations (Appendix A) yielded negative A_min values for 27 children and a negative BMI_min value for one child. In addition, one child had an estimate of BMIV_max that was deemed to be implausible (60.24 kg/m²) after his or her raw data were checked. These observations were deleted from the sample and were not used in further analyses. The mean and 95% CIs for the 5 BMI growth traits were plotted by sex and birth cohort.

In addition, to assess whether BMI trends were due to changes in weight, height, or both weight and height, each individual's observed weight and height (and BMI) data closest to their estimated A_min and also closest to their estimated AV_max were converted to z scores relative to the CDC 2000 reference (18), and separate ANOVA models (including a Bonferroni correction for multiple comparisons) for each dimension and each sex were used to test for differences between cohorts. The CDC reference was used, in this instance, because it includes data covering the entire age range necessary.

Univariable sex-specific general linear regression models were built to examine the unadjusted effect of each covariate on each BMI trait. Then, multivariable models for boys and girls separately were built to examine birth cohort effects on the BMI growth traits after adjustment for covariates, including maternal BMI and infant weight gain. The z scores for birth weight and weight at 2 y of age were calculated relative to the WHO 2006 Child Growth Standard (16), and infant weight gain from birth to 2 y was calculated as the residuals from sex-specific linear regressions of z scores at 2 y on z scores at birth (19). Covariate data were missing for some participants (n values in Table 1) so they were imputed 5 times by using ice in Stata. The imputation models included the 5 BMI growth traits, birth year, and all covariates (ie, sex, gestational age, birth order, birth weight z scores, conditional weight z scores at age 2 y, and maternal age, height, and BMI). The distributions of the imputed variables were similar to those of the observed data, which provided evidence that the missing data were missing at random. Micombine in Stata was used to simultaneously fit the regression models to each of the 5 imputation sets and estimate single-parameter estimates. The analyses were conducted by using Stata IC10 (StataCorp).

TABLE 1.

Description of the study sample, by birth cohort

	Missingdata	Cohort 1: 1928–1953 (n = 287)	Cohort 2: 1954–1972 (n = 297)	Cohort 3: 1973–1999 (n = 271)	P1	Total (n = 855)
Female [% (n)]	0	45.3 (130)	51.9 (154)	49.4 (134)	0.278	48.9 (418)
Birth year	0	1941 ± 7.432	1963 ± 5.49	1984 ± 6.88	<0.001	1962 ± 18.72
Gestational age at birth (wk)	359	39.73 ± 1.71	39.66 ± 1.84	39.93 ± 1.41	0.389	39.76 ± 1.67
Gestational age after imputation (wk)3	—	39.62 ± 1.74	39.71 ± 1.73	39.83 ± 1.61		39.72 ± 1.70
Birth weight (kg)	69	3.31 ± 0.56	3.36 ± 0.51	3.48 ± 0.55	0.001	3.38 ± 0.55
Birth weight (z score)4	69	−0.04 ± 1.24	0.10 ± 1.08	0.34 ± 1.16	0.001	0.12 ± 1.17
Birth weight after imputation (z score)	—	−0.05 ± 1.24	0.10 ± 1.08	0.35 ± 1.16		0.13 ± 1.17
Weight at 2 y (kg)	264	12.43 ± 1.32	12.20 ± 1.29	11.97 ± 1.33	0.008	12.26 ± 1.32
Weight at 2 y (z score)	264	0.34 ± 0.86	0.23 ± 0.88	0.02 ± 0.83	0.009	0.24 ± 0.87
Conditional weight at 2 y (z score)5	264	0.12 ± 0.79	0.00 ± 0.84	−0.31 ± 0.80	<0.001	0.00 ± 0.82
Conditional weight at 2 y after imputation (z score)	—	0.10 ± 0.81	-0.01 ± 0.85	-0.33 ± 0.80		–0.07 ± 0.84
Maternal age (y)	75	29.62 ± 6.04	26.35 ± 4.49	28.93 ± 5.42	<0.001	28.22 ± 5.52
Maternal age after imputation (y)	—	29.55 ± 5.99	26.35 ± 4.51	28.84 ± 5.40		28.21 ± 5.50
Maternal height (m)	53	163.64 ± 5.51	163.91 ± 5.68	165.45 ± 5.64	0.001	164.26 ± 5.65
Maternal height after imputation (m)	—	163.66 ± 5.50	163.92 ± 5.70	165.42 ± 5.68		164.31 ± 5.68
Maternal BMI (kg/m2)	64	22.65 ± 3.70	22.48 ± 3.33	23.01 ± 4.31	0.268	22.69 ± 3.76
Maternal BMI after imputation (kg/m2)	—	22.63 ± 3.71	22.48 ± 3.33	23.05 ± 4.18		22.71 ± 3.75

¹Differences between cohorts were tested by using ANOVA for continuous variables and chi-square test for categorical variables.

²Mean ± SD (all such values).

³Missing covariate data were imputed 5 times by using ice in Stata. The means over the 5 imputation runs are presented.

⁴The z scores were calculated relative to the WHO growth standard.

⁵Calculated as the residuals from the linear regression of z scores at 2 y on z scores at birth.

RESULTS

The mean birth year in the analysis sample was 1962, and weight at birth and at age 2 y were within 0.35 z scores of the median of the WHO standard (16) (Table 1). Mean birth weight increased across cohorts by 200 g (P < 0.001), and mean weight at 2 y decreased by 300 g (P = 0.008); thus, infants in the 1973–1999 cohort had slower infant weight gain than did infants in the earlier 2 cohorts (P < 0.001). Maternal BMI did not show a significant secular trend, but maternal height in the 1973–1999 cohort was significantly greater than that in the 1928–1953 and 1954–1972 cohorts (P < 0.05).

All of the growth curve parameters (intercept, age, age², and age³) were significant in the mixed-effects models. For boys, the main effect for the 1954–1972 cohort and also the cohort-by-age² and cohort-by-age³ interactions for this cohort were significant (Table 2). No significant cohort effects were observed in the model for girls. Sex-specific BMI distance curves are provided in Figures 1 and 2 to describe the results of the mixed-effects models. Separate ANOVAs were conducted to test cohort differences in predicted BMI every 2 y of age. The F statistic for the cohort effect was significant (P < 0.05) at all ages except during midchildhood (ages 6, 8, and 10 y in boys; ages 6 and 8 y in girls), as also shown in Figures 1 and 2. A negative secular trend was present early in life; BMI at 2 y was ∼0.5 kg/m² lower for boys and girls born in 1973–1999 than for boys and girls born in 1928–1953. Boys and girls in the 1973–1999 cohort remained smaller than those in the other 2 cohorts until ∼5 y of age and then showed a greater rate of BMI increase to achieve BMI values at 18 y of age that were >1 unit greater than those of their peers of the same sex in the other cohorts.

TABLE 2.

Mixed-effects growth models of serial BMI for 855 children aged 2–18 y from the Fels Longitudinal Study, testing for cohort effects

	Boys (n = 437)		Girls (n = 418)
	β (SE)	P	β (SE)	P
Fixed part
Base model
Constant	19.113 (0.224)	<0.001	19.342 (0.251)	<0.001
Age (decimal y)	−1.374 (0.073)	<0.001	−1.717 (0.085)	<0.001
Age2	0.158 (0.006)	<0.001	0.204 (0.007)	<0.001
Age3	−0.004 (0.000)	<0.001	−0.006 (0.000)	<0.001
Birth cohort
1928–1953 (referent)	—		—
1954–1972	−0.691 (0.325)	0.034	−0.227 (0.346)	0.511
1973–1999	−0.607 (0.370)	0.101	−0.711 (0.388)	0.067
Interaction effects
Birth cohort
1954–1972 × age	0.185 (0.107)	0.083	0.134 (0.117)	0.256
1954–1972 × age2	−0.025 (0.009)	0.005	−0.015 (0.010)	0.580
1954–1972 × age3	0.001 (0.000)	0.001	0.000 (0.000)	0.883
1973–1999 × age	0.060 (0.120)	0.619	0.098 (0.131)	0.458
1973–1999 × age2	0.002 (0.010)	0.834	0.008 (0.012)	0.480
1973–1999 × age3	−0.000 (0.000)	0.932	−0.000 (0.000)	0.302
Random part
SD (constant)	2.122 (0.093)		2.112 (0.097)
SD (age)	0.638 (0.025)		0.675 (0.027)
SD (age2)	0.029 (0.001)		0.031 (0.001)
SD (residual)	0.674 (0.006)		0.744 (0.007)
Log likelihood	−10305.1		−10507.9

FIGURE 1.

Mean BMI growth curves from 2 to 18 y of age, by birth cohort for 437 boys from the Fels Longitudinal Study, estimated from a mixed-effects model. P values were derived from separate ANOVAs that were applied to the estimated BMI data at each age to test for differences between cohorts. All ANOVAs included a multiple comparison (with a Bonferroni correction) to test the difference between any 2 cohorts, although only the P values for the overall models are shown here.

FIGURE 2.

Mean BMI growth curves from 2 to 18 y of age, by birth cohort for 418 girls from the Fels Longitudinal Study, estimated from a mixed-effects model. P values were derived from separate ANOVAs that were applied to the estimated BMI data at each age to test for differences between cohorts. All ANOVAs included a multiple comparison (with a Bonferroni correction) to test the difference between any 2 cohorts, although only the P values for the overall models are shown here.

Some differences between boys and girls in the relation of birth cohort to mean BMI at each age were observed. Girls in the 1954–1972 cohort had BMI values similar to those of girls in the 1928–1953 cohort at 2 y of age but then showed greater BMI gains and remained >0.5 units larger than girls in the 1928–1953 cohort after 6 y of age. Boys in the 1954–1972 cohort, conversely, were consistently ∼0.3 units smaller than were boys in the 1928–1953 cohort between 2 and 18 y of age.

To make better use of the longitudinal aspect of the BMI data, we estimated BMI growth curve traits for each child, including age at the minimum BMI, age at maximum BMI velocity, and others. In crude analyses, the mean and 95% CIs for the 5 BMI growth traits are plotted in Figure 3 by birth cohort for each sex separately. Nonoverlapping CIs indicate statistically significant group differences in the means. The differences between cohorts for A_min and BMI_min varied by sex; A_min decreased across cohorts for girls, from 1928–1953 to 1973–1999, but not for boys, and BMI_min, conversely, decreased across cohorts for boys but not for girls. The decreasing BMI_min across cohorts in boys was attributable to a decrease in weight rather than to an increase in height (see Supplementary Table 1 under “Supplemental data” in the online issue). That is, for boys the mean weight z score at A_min was 0.03 (SD: 0.83) for the 1928–1953 cohort compared with −0.20 (0.77) for the 1973–1999 cohort (P = 0.078); mean height z scores at A_min also decreased between the 1928–1953 and 1973–1999 cohorts, but only by ∼0.03 units (P = 0.999). The crude analyses of the individual estimates of the BMI growth traits in Figure 3 also showed that BMIV_max and V_max increased across cohorts in both sexes. Using the observed data, however, the increasing BMI at maximum BMI velocity across cohorts was observed only in girls and was more attributable to an increase in weight than to any secular change in height (see Supplementary Table 1 under “Supplemental data” in the online issue). That is, for girls the mean weight z scores at V_max increased by ∼0.3 units between the 1928–1953 and 1973–1999 cohorts (P = 0.015), and mean height z scores at V_max also increased by ∼0.2 units between these cohorts (P = 0.229).

FIGURE 3.

Crude mean and 95% CIs for 5 BMI growth traits, by birth cohort and sex. Nonoverlapping CIs indicate statistically significant group differences between the means. A_min, age at adiposity rebound; AV_max, age at maximum BMI velocity; BMI_min, BMI at adiposity rebound; BMIV_max, BMI at maximum BMI velocity; V_max, BMI velocity at maximum BMI velocity.

To assess whether cohort differences in BMI growth traits were due to differences in maternal BMI, infant weight gain, and other covariates, sex-specific multivariable regression models were produced (Tables 3 and 4). The unadjusted relation of each covariate with each BMI growth trait can be seen elsewhere (see Supplementary Table 2 under “Supplemental data” in the online issue). After adjustment for variations in infant weight gain, maternal BMI, and other covariates, A_min was still significantly lower in the 1973–1999 cohort and also in the 1954–1972 cohort, compared with the referent cohort (ie, 1928–1953) for girls, whereas no significant cohort effects on A_min were observed for boys. BMIV_max and V_max still increased significantly across cohorts for both boys and girls, with children in the 1973–1999 cohort being ∼1 unit larger and experiencing a 0.15-unit faster velocity per year (P < 0.01) compared with the referent cohort. For BMI_min, significant cohort effects were then observed in both boys and girls. AV_max was ∼1 y greater for boys in the 1954–1973 cohort than in the referent group, as in the crude analysis, but a previously unobserved significantly lower AV_max was now observed for girls in the 1973–1999 cohort. Also important to note is that children of mothers with higher BMI values had significantly earlier A_min and higher BMI_min and also significantly higher BMIV_max and faster V_max. These trends, apart from that for faster V_max, were also true of children who had greater weight gains as infants.

TABLE 3.

Multivariable general linear regression analysis of 5 BMI growth traits for 437 boys from the Fels Longitudinal Study¹

	β (SE)
	Adiposity rebound		Maximum BMI velocity
	Amin (n = 417)	BMImin (n = 436)	AVmax (n = 437)	BMIVmax (n = 436)	Vmax (n = 437)
	y	kg/m2	y	kg/m2	kg ⋅ m ⋅ yminus1
Gestational age (wk)	0.05 (0.07)	−0.05 (0.05)	0.04 (0.10)	−0.02 (0.22)	−0.01 (0.02)
Birth order
First (referent)	—	—	—	—	—
Second	0.49 (0.18)**	−0.12 (0.12)	0.68 (0.27)*	0.43 (0.43)	0.03 (0.04)
Third or greater	0.19 (0.20)	−0.06 (0.14)	0.26 (0.30)	0.75 (0.49)	0.07 (0.05)
Birth weight (z score)2	−0.03 (0.09)	0.12 (0.06)*	−0.12 (0.12)	−0.02 (0.21)	−0.01 (0.02)
Conditional weight at 2 y (z score)3	−0.27 (0.10)**	0.51 (0.07)***	−0.36 (0.15)*	0.85 (0.26)**	0.03 (0.02)
Maternal age (y)	0.02 (0.02)	−0.01 (0.01)	−0.02 (0.02)	−0.08 (0.04)*	−0.01 (0.00)*
Maternal height (m)	0.01 (0.01)	−0.01 (0.01)	−0.00 (0.02)	−0.04 (0.03)	−0.00 (0.00)
Maternal BMI (kg/m2)	−0.09 (0.02)***	0.05 (0.01)***	0.02 (0.03)	0.30 (0.05)***	0.03 (0.00)***
Birth cohort
1928–1953 (referent)	—	—	—	—	—
1954–1972	0.26 (0.19)	−0.39 (0.13)**	1.02 (0.27)***	0.36 (0.43)	0.01 (0.04)
1973–1999	−0.14 (0.19)	−0.17 (0.13)	0.22 (0.28)	1.36 (0.45)**	0.16 (0.04)***
Adjusted R2 range4	0.07–0.10	0.20–0.24	0.05–0.08	0.11–0.12	0.10–0.12

¹Micombine in Stata was used to apply the models to the 5 imputation data sets and to estimate single-parameter estimates. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001 (2 tailed). A_min, age at adiposity rebound; AV_max, age at maximum BMI velocity; BMI_min, BMI at adiposity rebound; BMIV_max, BMI at maximum BMI velocity; V_max, BMI velocity at maximum BMI velocity.

²The z scores were calculated relative to the WHO growth standard.

³Calculated as the residuals from the linear regression of z scores at 2 y on z scores at birth.

⁴Micombine in Stata does not produce a single r² value, so the range from the models applied to the 5 imputation data sets is provided.

TABLE 4.

Multivariable general linear regression analysis of 5 BMI growth traits for 418 girls from the Fels Longitudinal Study¹

	β (SE)
	Adiposity rebound		Maximum BMI velocity
	Amin (n = 411)	BMImin (n = 418)	AVmax (n = 418)	BMIVmax (n = 418)	Vmax (n = 418)
	y	kg/m2	y	kg/m2	kg ⋅ m ⋅ yminus1
Gestational age (wk)	0.02 (0.07)	−0.00 (0.04)	0.04 (0.07)	0.01 (0.08)	0.00 (0.01)
Birth order
First (referent)	—	—	—	—	—
Second	−0.20 (0.19)	0.07 (0.13)	−0.03 (0.19)	0.23 (0.29)	0.03 (0.03)
Third or greater	0.06 (0.21)	−0.22 (0.13)	0.02 (0.21)	−0.17 (0.31)	0.01 (0.04)
Birth weight (z score)2	−0.10 (0.09)	0.15 (0.05)**	−0.12 (0.08)	0.13 (0.11)	−0.00 (0.01)
Conditional weight at 2 y (z score)3	−0.47 (0.10)***	0.76 (0.07)***	−0.37 (0.11)**	1.03 (0.16)***	0.03 (0.02)
Maternal age (y)	0.00 (0.02)	0.02 (0.01)	−0.00 (0.02)	0.01 (0.02)	−0.00 (0.00)
Maternal height (m)	0.03 (0.02)*	−0.02 (0.01)*	0.01 (0.02)	−0.06 (0.02)**	−0.01 (0.00)*
Maternal BMI (kg/m2)	−0.09 (0.02)***	0.05 (0.01)**	−0.02 (0.02)	0.20 (0.03)***	0.02 (0.00)***
Birth cohort
1928–1953 (referent)	—	—	—	—	—
1954–1972	−0.40 (0.19)*	0.38 (0.11)**	−0.24 (0.19)	0.56 (0.28)*	0.03 (0.03)
1973–1999	−0.80 (0.21)***	0.08 (0.13)	−0.42 (0.21)*	0.91 (0.30)**	0.14 (0.03)***
Adjusted R2 range4	0.10–0.14	0.35–0.41	0.03–0.04	0.20–0.24	0.12–0.13

¹Micombine in Stata was used to apply the models to the 5 imputation data sets and to estimate single-parameter estimates. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001 (2 tailed). A_min, age at adiposity rebound; AV_max, age at maximum BMI velocity; BMI_min, BMI at adiposity rebound; BMIV_max, BMI at maximum BMI velocity; V_max, BMI velocity at maximum BMI velocity.

²The z scores were calculated relative to the WHO growth standards.

³Calculated as the residuals from the linear regression of z scores at 2 y on z scores at birth.

⁴Micombine in Stata does not produce a single r² value so the range from the models applied to the 5 imputation data sets is provided.

DISCUSSION

This article presents secular trends in BMI growth of European-American children over a 70-y period in the 20th century. The BMI growth of the most recent birth cohort (1973–1999) was different from that of children in the previous cohorts (1928–1953, 1954–1972). The observed trends reflect the same time course of accelerated BMI gain after 1960 seen in the US pediatric population (4, 20). A trend in increasing BMI may have started earlier in the century for black girls, but for European-American children (and black boys) the trend emerged in the 1960s (3). Few historical longitudinal data sets exist with which to compare our findings, although Himes (21) provides a comprehensive review. Using data from 22 studies in the United States between 1879 and 1970—including studies in Illinois (22), Ohio (23), and Pennsylvania (24)—he showed that when mean weights were plotted against mean statures on developmental reference charts (25), little evidence of a secular trend in the relation between weight and stature was found. Findings showing that the trend in increasing BMI only emerged during the 1960s have also been reported in studies of children in Europe (7, 9, 26).

In a Fels Longitudinal Study publication in 1991, no evidence of a secular trend in childhood BMI was found (27). However, in a more recent study of the population published in 2004, a significant secular trend in the rate of BMI increase after menarche was observed, so that by 6 y after menarche, the BMI for girls born in 1965–1983 was ∼2 units higher than that for girls born in 1929–1946 or 1947–1964 (28). At 6 y before menarche, the trend was much smaller, and girls born in 1965–1983 had BMI values similar to those of the other cohorts. The current report expands on previous publications and indicates that shifts in characteristics of the BMI growth curve have occurred starting after 1973 in boys and girls.

A negative secular trend in BMI was present at 2 y of age, with the most recent 1973–1999 cohort being ∼0.5 units smaller than children in the 1928–1953 cohort. Even though an extensive body of literature shows that individuals are developing greater BMI values at increasingly younger ages in childhood (5, 7, 10), the secular trend in BMI at the end of infancy may still be negative. The increased risk of childhood obesity in recent cohorts of individuals who demonstrated rapid infant weight gain (29, 30) was not studied here, and the findings of the current study do not necessarily suggest that this is not also a characteristic of Fels Longitudinal Study children.

In the study that first described the adiposity rebound, Rolland-Cachera et al (31) showed that BMI at age 1 y was significantly lower in those who had early rebound. Our findings corroborate and extend this observation, showing that children born more recently had a lower BMI at age 2 y, an earlier age at adiposity rebound (girls only), and subsequently a greater BMI growth velocity and BMI at the age of maximum velocity. Our group previously showed that early adiposity rebound is associated with the likelihood of overweight in adulthood (17), although it has been argued that the adiposity rebound is not a critical event in obesity development, but an epiphenomenon stemming from high BMI and/or upward centile crossing at the time of rebound being associated with earlier rebound and later obesity (32, 33). The current study indicates that early rebound followed by accelerated BMI gain may, at least in girls, occur in the absence of elevated concurrent or earlier BMI.

Although it should not alter the interpretation of the trends reported in the current study, the hypothesis that the adiposity rebound and the shape of the childhood BMI curve is a statistical artifact of the index should not be discounted. The BMI uses height squared when the appropriate power (to create an index that is independent of height) is closer to 3 in infancy and in adolescence (34), and this theoretically contributes to the U-shape of the curve. The influence of using height squared regardless of age on the shape of the curve warrants research and will help improve our understanding of the phenomenon of the adiposity rebound.

After the nadir in the BMI growth curve, children born in 1973–1999 showed accelerated BMI growth, and this trend amplified with age. The finding that the 1954–1972 cohort had lower levels of BMI gain compared with the 1928–1953 cohort for boys but not for girls was unexpected and has not been previously reported. The earlier ages of rebound and maximum BMI velocity of girls than of boys can be seen in the BMI curves. It is well known that pubertal maturation events occur earlier for girls (35, 36), but these results show that the entire BMI curve is shifted left on the age axis for girls compared with boys.

BMI is typically considered an indicator of adiposity rather than a growth trait per se. Little information, therefore, exists on whether the BMI growth curve traits discussed in the report can be profitably assessed as “developmental landmarks” and whether they reflect the maturation of the adipose tissue organ, similar to those documented for linear growth (37, 38), or are closely linked with the process of sexual maturation (39, 40). In the article that proposed the 5 BMI traits (17), V_max was significantly associated with adulthood (age 35–45 y) BMI in men and women, AV_max was associated with adulthood BMI in women but not in men, and BMIV_max was associated with adulthood BMI, total body fat, and percentage fat in men and with BMI in women. These growth traits, therefore, help characterize the trajectory of individuals on a course toward obesity and excess adiposity. However, further work is needed to discern whether the traits are associated with disease risk and also whether trends in total body fat and specific fat depots have occurred in tandem with the trends in BMI.

A large body of literature suggests a positive association of infant weight gain and maternal BMI with obesity risk (29, 41). Similarly, in the current study, greater infant weight gain and maternal BMI contributed to a more obesity-prone profile of BMI growth. These covariates did not, however, explain the cohort differences in BMI growth traits in regression analyses. The significant cohort effects observed in regression analyses that were not present in the crude analyses most likely reflect the fact that adjustment for covariates will have increased the precision of the parameter estimates (ie, narrower CIs) and therefore improved our ability to detect significant results. Maternal smoking and alcohol use during pregnancy, infant feeding practices, child physical activity and diet, and many other factors have likely changed over the period studied but have not been uniformly assessed over the duration of the Fels Longitudinal Study, which made it impossible to adjust for these variables in our analyses. Other limitations included a sample composed exclusively of European-American children born in southwestern Ohio, which limited the generalizability of the results to other ethnic groups, and available data ended with those born in 1999; thus, BMI trends of children born during the period of the national obesity epidemic were not fully captured. In addition, consistent “prepregnancy” maternal BMI and height data were not available for all subjects; therefore, “nonpregnant” data were used instead. Also, the dependent variables were calculated by using estimated data, so the SEs of the parameter estimates in the regression models may have been underestimated, which would increase the risk of type 1 error.

Our conclusion is that shifts in the BMI growth curve have occurred after the 1970s and are present in early childhood. Boys and girls in the most recent cohort had greater maximum BMI velocities and adolescent BMI values, but did not have higher BMI values at the age of adiposity rebound, compared with boys and girls of the same sex in the previous cohorts. Maternal BMI and infant weight gain were associated with an obesity-prone profile of BMI growth but did not mediate the cohort differences. Secular trends in childhood BMI are not increasingly monotonic at all ages, and pediatricians should be aware that lower BMI values before the adiposity rebound are characteristic of children born during the obesity epidemic, who subsequently showed rapid BMI gain.

APPENDIX A